Flow quantification in ultrasound using conditional random fields with global consistency

Abstract

Velocities are unaliased using conditional random fields. To constrain the energy minimization function, a global term includes a measure of a level of aliasing. In one example, the measure of the level of aliasing is based on a change in volume, such as the volume of the left ventricle. The unaliasing is performed along one or more surfaces, such as surfaces intersecting the mitral annulus and the left ventricle outflow tract. The anatomy used is identified and/or tracked using one or more machine-learnt detectors. Both B-mode and velocity information may be used for detecting the anatomy.

We claim: 
     
       1. A method for flow quantification in ultrasound imaging, the method comprising:
 acquiring, with a transducer, ultrasound data representing a cardiac region of a patient, the ultrasound data comprising velocity values of flow of fluid in the cardiac region of the patient; 
 calculating, with a processor of an ultrasound imaging system, a volume change from the ultrasound data; 
 unaliasing, with the processor of the ultrasound imaging system, the velocity values as a function of the volume change and a conditional random field; 
 calculating, with the processor of the ultrasound imaging system, a quantity as a function of the unaliased velocity values; and 
 displaying, on a display, the quantity. 
 
     
     
       2. The method of  claim 1  wherein acquiring comprises acquiring the velocity values for the cardiac region, where the cardiac region comprises a volume of the heart. 
     
     
       3. The method of  claim 1  wherein acquiring comprises also acquiring B-mode data representing the cardiac region at different times, and wherein calculating the volume change comprises calculating the volume change with the B-mode data. 
     
     
       4. The method of  claim 3  wherein calculating the volume change comprises calculating a left ventricle volume change. 
     
     
       5. The method of  claim 1  wherein unaliasing comprises estimating a number of aliased velocities, and including a global constraint in an energy function conditional random field, the global constraint being a function of the number. 
     
     
       6. The method of  claim 1  wherein unaliasing comprises determining a set of velocities with a minimum energy of the conditional random field. 
     
     
       7. The method of  claim 6  wherein an energy of the conditional random field comprises a data term that is a function of the velocity values, a spatial pair-wise term, and a global term. 
     
     
       8. The method of  claim 1  wherein unaliasing comprises unaliasing the velocity values along one or more surfaces for an inflow, an outflow, or both inflow and outflow regions. 
     
     
       9. The method of  claim 1  wherein calculating comprises calculating a volume flow. 
     
     
       10. The method of  claim 1  wherein displaying comprises displaying the quantity with an image of the cardiac region. 
     
     
       11. The method of  claim 1  wherein acquiring comprises acquiring B-mode information;
 further comprising:
 detecting a position, orientation, and scale of a left ventricle boundary with a machine-trained marginal space detector; 
 estimating a mitral annulus and left ventricle outflow tract with a machine-trained probabilistic boosting tree classifier, the probabilistic boosting tree classifier using a local search constrained by the detected position, orientation, and scale of the left ventricle boundary; 
 wherein calculating comprises calculating as a function of the left ventricle boundary; and 
 wherein unaliasing comprises unaliasing in a first surface intersecting the mitral annulus and in a second surface intersecting the left ventricle outflow tract. 
 
 
     
     
       12. The method of  claim 11  further comprising:
 refining the orientation as a function of the velocity values. 
 
     
     
       13. The method of  claim 1  further comprising:
 detecting a left ventricle boundary, a mitral annulus, and a left ventricle outflow track; 
 tracking the left ventricle boundary, mitral annulus and left ventricle outflow track over time with a Bayesian network of local features and plane position.